An artificial intelligence-based non-invasive cuff blood pressure monitoring method and related devices

Abstract

The application discloses a non-invasive cuff blood pressure monitoring method based on artificial intelligence and related equipment, which obtains multi-modal physiological signals such as cuff inflation pressure, cuff interface pressure, pulse wave and electrocardio by means of quality assessment and adaptive weighted fusion of the signals, so as to obtain a high-reliability fusion signal. The deep learning model is used for feature extraction and time sequence modeling of the oscillation wave envelope, and the blood pressure parameter prediction result is output. The individual blood pressure baseline model is constructed in combination with the historical measurement data, the prediction result is corrected, and the alarm threshold is dynamically generated, so that the early warning of blood pressure abnormalities is realized. Through real-time comparison of multi-modal pressure signals, safety arbitration and abnormal protection control are performed, inflation is terminated or pressure relief operation is performed in time when an abnormality is detected, and a degraded monitoring mode is supported. The method can effectively improve the accuracy, stability and safety of blood pressure measurement, and has good clinical application value.

Description

Technical Field

[0001] This specification relates to the field of health monitoring, and more specifically, this application relates to a non-invasive cuff-type blood pressure monitoring method and related equipment based on artificial intelligence. Background Technology

[0002] In preclinical practice, noninvasive blood pressure monitoring primarily employed cuff-based measurement methods based on the oscillatory wave principle. These methods analyze the pressure oscillation signals generated during the gradual inflation and deflation of the cuff to determine systolic, diastolic, and mean arterial pressure. While this type of method is simple in structure and widely used, it still has several shortcomings in practical applications.

[0003] On the one hand, traditional systems typically rely on a single pressure monitoring channel, making them susceptible to factors such as cuff tightness, size matching, and airway sealing. When signal distortion or sensor drift occurs, it is difficult to detect and correct it in a timely manner, leading to measurement errors and even safety risks. On the other hand, existing methods mostly use fixed algorithm models and empirical thresholds for blood pressure calculation and alarm determination, lacking the ability to adapt to different individual physiological differences and failing to reflect the trend of blood pressure changes relative to the patient's baseline.

[0004] Furthermore, in situations involving motion interference, arrhythmias, or significant signal noise, the traditional oscillatory wave method has limited ability to identify abnormal waveforms, easily leading to unstable or inaccurate measurement results. Simultaneously, existing systems typically rely on manually set alarm thresholds, which cannot be dynamically adjusted based on the patient's condition, resulting in missed or false alarms, increasing the clinical burden and impacting monitoring efficiency.

[0005] Therefore, it is necessary to propose a non-invasive cuff-type blood pressure monitoring method and related equipment based on artificial intelligence to at least solve some of the above-mentioned problems. Summary of the Invention

[0006] The summary section introduces a series of simplified concepts, which will be further explained in detail in the detailed description section. This summary section is not intended to limit the key and essential technical features of the claimed technical solution, nor is it intended to determine the scope of protection of the claimed technical solution.

[0007] Firstly, this application proposes a non-invasive cuff-based blood pressure monitoring method based on artificial intelligence, comprising: Acquire multimodal physiological signal data, wherein the multimodal physiological signal data includes at least cuff inflation pressure signal, cuff interface pressure signal, pulse wave signal and electrocardiogram signal; The signal quality of the above multimodal physiological signal data was evaluated to obtain the quality level of each modality signal; Based on the above signal quality levels, the above multimodal physiological signal data are adaptively weighted and fused to obtain a fused signal; Based on a deep learning model, feature extraction and time-series modeling are performed on the oscillatory wave envelope in the above fused signal to obtain blood pressure parameter prediction results. An individualized blood pressure baseline model is constructed based on historical measurement data, and the prediction results of the above blood pressure parameters are corrected according to the above baseline model; Based on the above blood pressure parameter prediction results and the above baseline model, alarm thresholds are dynamically generated, and abnormal blood pressure warnings are executed. Safety arbitration and anomaly protection control are performed based on multimodal pressure signals to complete the safety management of the blood pressure measurement process.

[0008] In one feasible implementation, the above-mentioned signal quality assessment of the multimodal physiological signal data to obtain the quality level of each modality signal includes: Quality assessment features of physiological signals of each modality were extracted, including signal amplitude stability, signal-to-noise ratio, waveform integrity and motion interference index. Based on the above quality evaluation characteristics, a signal quality assessment index is constructed. The above signal quality evaluation indicators are compared with preset quality thresholds to obtain the corresponding signal quality level.

[0009] In one feasible implementation, the above-mentioned adaptive weighted fusion of the multimodal physiological signal data based on the above-mentioned signal quality level to obtain a fused signal includes: Assign corresponding fusion weights based on the quality level of each modal signal; Based on the above fusion weights, the modal signals are weighted and calculated to obtain the initial fused signal; When the quality of the first modal signal is detected to be lower than a preset threshold and the quality of the second modal signal is higher than a preset threshold, the fusion weight is adjusted so that the second modal signal becomes the dominant signal. When the quality of all multimodal signals is below a preset threshold, the current measurement is terminated and a signal abnormality prompt is output.

[0010] In one feasible implementation, the aforementioned feature extraction and temporal modeling of the oscillatory wave envelope in the fused signal based on a deep learning model yields blood pressure parameter prediction results, including... Based on the convolutional neural network, the morphological features of the envelope of the above oscillating wave are extracted to obtain the oscillating wave feature vector, which includes the rising branch slope, the falling branch slope, the peak position, the envelope width and the asymmetry index. The oscillation wave feature vector is modeled temporally using a long short-term memory network to obtain the dynamic change law of the oscillation wave, and abnormal oscillation waveforms are identified and eliminated to obtain blood pressure parameter prediction results.

[0011] In one feasible implementation, the above-mentioned construction of an individualized blood pressure baseline model based on historical measurement data, and the correction of the blood pressure parameter prediction results based on the baseline model, includes: Acquire historical blood pressure measurement data and corresponding multimodal feature data within a preset time window; Outlier removal and data cleaning were performed on the aforementioned historical measurement data. Calculate the baseline values ​​for systolic blood pressure, diastolic blood pressure, and heart rate; An individualized blood pressure baseline model was constructed based on the above baseline values; The bias of the current blood pressure parameter prediction results is corrected based on the above-mentioned individualized blood pressure baseline model.

[0012] In one feasible implementation, the above-mentioned dynamic generation of alarm thresholds based on the blood pressure parameter prediction results and the baseline model, and the execution of abnormal blood pressure warnings, includes: The static alarm thresholds are calculated based on the above individualized blood pressure baseline model, whereby the static alarm thresholds include the upper limit threshold for systolic blood pressure and the lower limit threshold for systolic blood pressure. Based on the above blood pressure parameter prediction results, a short-term blood pressure change trend model is constructed to predict the blood pressure change trend within a preset time period. Calculate the relative change threshold based on the above blood pressure change trend; When the predicted blood pressure parameters exceed the static alarm threshold or the relative change threshold, a blood pressure abnormality warning is triggered.

[0013] In one feasible implementation, the above-mentioned safety arbitration and anomaly protection control based on multimodal pressure signals to complete the safety management of the blood pressure measurement process includes: The cuff inflation pressure signal and the cuff interface pressure signal are collected in real time, and the current cuff pressure value corresponding to the above cuff inflation pressure signal is displayed; at the same time, the above cuff inflation pressure signal and the above cuff interface pressure signal are compared in real time, and the pressure difference between the two is calculated. When the pressure difference exceeds the preset difference threshold, or when the cuff pressure exceeds the preset maximum pressure threshold and the duration exceeds the set duration, it is determined to be an abnormal state. In the above-mentioned abnormal state, safety control operations are performed, including at least triggering the pressure protection fuse mechanism, terminating inflation, opening the pressure relief valve for automatic pressure relief, and stopping the current measurement process.

[0014] Secondly, the present invention also proposes an artificial intelligence-based non-invasive cuff blood pressure monitoring system for implementing the artificial intelligence-based non-invasive cuff blood pressure monitoring method as described in any one of the first aspects of the claim, comprising: The acquisition unit is used to acquire multimodal physiological signal data, wherein the multimodal physiological signal data includes at least cuff inflation pressure signal, cuff interface pressure signal, pulse wave signal and electrocardiogram signal; The evaluation unit is used to evaluate the signal quality of the above multimodal physiological signal data and obtain the quality level of each modality signal. The fusion unit is used to adaptively weight and fuse the multimodal physiological signal data based on the above signal quality levels to obtain a fused signal; The modeling unit is used to extract features and perform temporal modeling of the oscillatory wave envelope in the above-mentioned fused signal based on a deep learning model, so as to obtain blood pressure parameter prediction results. The calibration unit is used to construct an individualized blood pressure baseline model based on historical measurement data, and to correct the blood pressure parameter prediction results based on the baseline model. The generation unit is used to dynamically generate alarm thresholds based on the above blood pressure parameter prediction results and the above baseline model, and to perform blood pressure abnormality warnings. The safety management unit is used to perform safety arbitration and anomaly protection control based on multimodal pressure signals to complete the safety management of the blood pressure measurement process.

[0015] Thirdly, the present invention also proposes an electronic device comprising: a memory and a processor, characterized in that the processor is used to execute a computer program stored in the memory to implement the steps of the artificial intelligence-based non-invasive cuff blood pressure monitoring method as described in any of the first aspects.

[0016] Fourthly, the present invention also proposes a computer-readable storage medium having a computer program stored thereon, characterized in that, when the computer program is executed by a processor, it implements the steps of the artificial intelligence-based non-invasive cuff blood pressure monitoring method as described in any of the first aspects.

[0017] In summary, this invention constructs a multi-dimensional perception and intelligent analysis system by introducing multimodal physiological signal fusion and artificial intelligence processing mechanisms, achieving comprehensive optimization of the blood pressure measurement process. By fusing cuff inflation pressure, interface pressure, pulse wave, and ECG signals, and combining signal quality assessment with adaptive weighted processing, the stability and reliability of measurements under complex interference environments are improved. This invention significantly improves blood pressure calculation accuracy and enhances the ability to identify abnormal waveforms by using a deep learning model to extract features and perform time-series modeling of the oscillating wave envelope. Based on historical data, this invention constructs an individualized blood pressure baseline model and dynamically generates alarm thresholds, realizing a shift from fixed thresholds to individualized and trend-based early warnings, which helps to identify abnormal blood pressure changes in advance. Through a multimodal pressure signal safety arbitration and anomaly protection mechanism, this invention effectively reduces safety risks caused by sensor failure or abnormal operating conditions, and supports degraded monitoring operation when necessary, thereby improving the overall intelligence level, measurement accuracy, and safety of the system.

[0018] Other advantages, objectives and features of this application will be apparent in part from the description which follows, and in part from what those skilled in the art will understand through study and practice of this application. Attached Figure Description

[0019] Various other advantages and benefits will become apparent to those skilled in the art upon reading the following detailed description of preferred embodiments. The accompanying drawings are for illustrative purposes only and are not intended to limit this specification. Furthermore, the same reference numerals denote the same parts throughout the drawings. In the drawings: Figure 1 A schematic flowchart of a non-invasive cuff-type blood pressure monitoring method based on artificial intelligence, provided for an embodiment of this application; Figure 2 A structural schematic diagram of a non-invasive cuff-type blood pressure monitoring system based on artificial intelligence, provided for an embodiment of this application; Figure 3 This is a schematic diagram of an electronic device structure provided in an embodiment of this application. Detailed Implementation

[0020] The terms "first," "second," "third," "fourth," etc. (if present) in the specification, claims, and accompanying drawings of this application are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments described herein can be implemented in a sequence other than that illustrated or described herein. Furthermore, the terms "comprising" and "having," and any variations thereof, are intended to cover non-exclusive inclusion; for example, a process, method, system, product, or apparatus that comprises a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or apparatus. The technical solutions of the embodiments of this application will now be clearly and completely described in conjunction with the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of this application, and not all of them.

[0021] Please see Figure 1 This is a flowchart illustrating a non-invasive cuff-type blood pressure monitoring method based on artificial intelligence, provided in an embodiment of this application. Specifically, it may include: S110. Acquire multimodal physiological signal data, wherein the multimodal physiological signal data includes at least cuff inflation pressure signal, cuff interface pressure signal, pulse wave signal and electrocardiogram signal; S120. The signal quality of the above multimodal physiological signal data is evaluated to obtain the quality level of each modality signal; S130. Based on the above signal quality levels, the above multimodal physiological signal data are adaptively weighted and fused to obtain a fused signal; S140. Based on a deep learning model, feature extraction and time-series modeling are performed on the oscillatory wave envelope in the above fused signal to obtain blood pressure parameter prediction results. S150. Construct an individualized blood pressure baseline model based on historical measurement data, and correct the blood pressure parameter prediction results based on the baseline model. S160. Based on the above blood pressure parameter prediction results and the above baseline model, dynamically generate alarm thresholds and execute blood pressure abnormality warnings; S170: Performs safety arbitration and anomaly protection control based on multimodal pressure signals to complete the safety management of the blood pressure measurement process.

[0022] For example, in step S110, an inflation pressure monitoring unit installed on the main airway of the cuff acquires the airway pressure signal in real time during the inflation and deflation of the cuff, serving as the basic data source for blood pressure measurement. Simultaneously, a flexible pressure sensor deployed on the inner wall of the cuff bladder or the contact surface between the bladder and the patient's skin acquires the cuff interface pressure signal to reflect the actual pressure state between the cuff and the limb. Further, a photoplethysmography (PPG) sensor installed in the inner layer of the cuff acquires the patient's pulse wave signal, extracting pulse waveform characteristics and pulse wave conduction time-related parameters. Electrocardiogram (ECG) signals are acquired using at least two ECG electrodes to obtain heart rate and ECG rhythm information, thereby forming multimodal physiological signal data encompassing mechanical pressure information and physiological electrical signals, providing a multidimensional data foundation for subsequent blood pressure calculation and safety control.

[0023] In step S120, the signal quality of the above multimodal physiological signal data is evaluated. The quality evaluation characteristics of each modality signal are calculated, including signal amplitude stability, waveform continuity, signal-to-noise ratio, degree of abnormal fluctuation, and intensity of motion interference reflected by acceleration or waveform changes. Based on the preset quality evaluation model, each modality signal is scored and classified into excellent, good, or poor quality levels, thereby identifying the reliability and usability of the current signal and providing a basis for subsequent fusion processing.

[0024] In step S130, fusion weights are dynamically allocated based on the quality level of each modal signal. Adaptive weighted fusion processing is performed on the multimodal physiological signal data, prioritizing higher-quality modal signals as the dominant data source, while reducing the weight of lower-quality modal signals or suppressing them. When a significant decrease in the quality of a key modal signal is detected, the system automatically switches to a backup modality or redundant channel to obtain a fusion signal with stronger anti-interference capabilities, ensuring the stability and robustness of blood pressure calculation.

[0025] In step S140, feature extraction and temporal modeling are performed on the oscillatory wave envelope in the fused signal based on a deep learning model. Specifically, a convolutional neural network is used to extract morphological features of the oscillatory wave envelope, including the slope of the rising branch, the slope of the falling branch, the peak position, the envelope width, and asymmetry indices. Furthermore, a long short-term memory network is used to model the temporal variation of the oscillatory wave envelope during the deflation process to identify normal and abnormal waveforms (such as waveform distortion caused by arrhythmias), and abnormal oscillatory waves are screened out or corrected. Finally, the predicted results for systolic blood pressure, diastolic blood pressure, and mean arterial pressure are output.

[0026] In step S150, an individualized blood pressure baseline model is constructed based on historical multiple measurement data. This involves cleaning and removing outliers from historical blood pressure data within a preset time window, calculating the patient's baseline systolic blood pressure, baseline diastolic blood pressure, and baseline heart rate, and then combining this with multimodal features to build the individualized blood pressure model. Based on this, the current blood pressure prediction result is corrected for deviations according to the baseline model, ensuring that the measurement results not only conform to instantaneous measurement characteristics but also reflect the patient's long-term physiological state, thereby improving measurement accuracy and clinical reference value.

[0027] In step S160, an alarm threshold is dynamically generated based on the blood pressure parameter prediction results and the individualized baseline model. A baseline-based relative change threshold is introduced on the basis of the traditional absolute threshold, and a prediction model is constructed in combination with the short-term blood pressure change trend to predict blood pressure changes in the future. When the prediction results indicate that the blood pressure is about to exceed the set threshold or shows a continuous upward or downward trend, an early warning of abnormal blood pressure is triggered, thereby realizing the early identification of the risk of hypotension or hypertension and providing a time window for clinical intervention.

[0028] Finally, in step S170, safety arbitration and abnormal protection control are performed based on the multimodal pressure signal. By comparing the difference between the cuff inflation pressure signal and the cuff interface pressure signal in real time, if the difference exceeds a preset threshold and the duration exceeds a set time, it is determined that there is an abnormality in the air path, cuff loosening, or sensor failure, and a safety control strategy is immediately triggered, including terminating inflation, opening the deflation valve for rapid pressure relief, and stopping the current measurement process. At the same time, when an abnormality or signal failure of the pressure sensor is detected, the system automatically switches to a continuous monitoring mode based on pulse wave and electrocardiogram signals. Blood pressure trend estimation is achieved through parameters such as pulse wave conduction time to ensure that basic monitoring functions can still be maintained under abnormal conditions, thereby realizing intelligent, safe, and continuous management of the entire process of non-invasive cuff blood pressure monitoring.

[0029] In one feasible implementation, the above-mentioned signal quality assessment of the multimodal physiological signal data to obtain the quality level of each modality signal includes: Quality assessment features of each modality of physiological signal were extracted, including signal amplitude stability, signal-to-noise ratio, waveform integrity and motion interference index. Based on the above quality evaluation characteristics, a signal quality assessment index is constructed. The above signal quality evaluation indicators are compared with preset quality thresholds to obtain the corresponding signal quality level.

[0030] For example, the quality assessment process for multimodal physiological signal data first involves feature extraction for different signal sources. For cuff inflation pressure signals and cuff interface pressure signals, the focus is on analyzing the pressure change curves during inflation and deflation. By calculating the amplitude fluctuation range of the pressure waveform, the consistency of changes between adjacent sampling points, and the continuity of the oscillation envelope, characteristic parameters reflecting signal amplitude stability and waveform integrity are obtained. For pulse wave signals, the signal-to-noise ratio and waveform clarity are evaluated by analyzing the rising edge, falling edge, dicrotic wave structure, and perfusion index changes. For electrocardiogram (ECG) signals, signal quality is reflected by detecting the accuracy of R-wave peak recognition, RR interval stability, and baseline drift. Simultaneously, motion interference indicators are extracted based on the motion interference reflected by abnormal fluctuations in pressure waveforms, pulse wave distortion, and irregular changes in ECG signals.

[0031] After completing the multi-dimensional feature extraction, the various quality evaluation features are further normalized and weighted according to preset weights to construct a unified signal quality evaluation index, which comprehensively reflects the overall reliability of the current signals across different modalities. Subsequently, the signal quality evaluation index is compared with a pre-set quality threshold. An index above the high-quality threshold is classified as excellent, in the middle range as good, and below the low-quality threshold as poor. This method enables real-time and objective evaluation of the quality of multimodal physiological signals, providing a reliable data foundation for subsequent multimodal fusion calculations, blood pressure parameter prediction, and anomaly handling, thereby effectively reducing the impact of motion interference, signal attenuation, and sensor malfunctions on blood pressure measurement results.

[0032] In one feasible implementation, the above-mentioned adaptive weighted fusion of the multimodal physiological signal data based on the above-mentioned signal quality level to obtain a fused signal includes: Assign corresponding fusion weights based on the quality level of each modal signal; Based on the above fusion weights, the modal signals are weighted and calculated to obtain the initial fused signal; When the quality of the first modal signal is detected to be lower than a preset threshold and the quality of the second modal signal is higher than a preset threshold, the fusion weight is adjusted so that the second modal signal becomes the dominant signal. When the quality of all multimodal signals is below a preset threshold, the current measurement is terminated and a signal abnormality prompt is output.

[0033] In one feasible implementation, the aforementioned feature extraction and temporal modeling of the oscillatory wave envelope in the fused signal based on a deep learning model yields blood pressure parameter prediction results, including... Based on the convolutional neural network, the morphological features of the envelope of the above oscillating wave are extracted to obtain the oscillating wave feature vector, which includes the rising branch slope, the falling branch slope, the peak position, the envelope width and the asymmetry index. The oscillation wave feature vector is modeled temporally using a long short-term memory network to obtain the dynamic change law of the oscillation wave, and abnormal oscillation waveforms are identified and eliminated to obtain blood pressure parameter prediction results.

[0034] In one feasible implementation, the above-mentioned construction of an individualized blood pressure baseline model based on historical measurement data, and the correction of the blood pressure parameter prediction results based on the baseline model, includes: Acquire historical blood pressure measurement data and corresponding multimodal feature data within a preset time window; Outlier removal and data cleaning were performed on the aforementioned historical measurement data. Calculate the baseline values ​​for systolic blood pressure, diastolic blood pressure, and heart rate; An individualized blood pressure baseline model was constructed based on the above baseline values; The bias of the current blood pressure parameter prediction results is corrected based on the above-mentioned individualized blood pressure baseline model.

[0035] In one feasible implementation, the above-mentioned dynamic generation of alarm thresholds based on the blood pressure parameter prediction results and the baseline model, and the execution of abnormal blood pressure warnings, includes: The static alarm thresholds are calculated based on the above individualized blood pressure baseline model, whereby the static alarm thresholds include the upper limit threshold for systolic blood pressure and the lower limit threshold for systolic blood pressure. Based on the above blood pressure parameter prediction results, a short-term blood pressure change trend model is constructed to predict the blood pressure change trend within a preset time period. Calculate the relative change threshold based on the above blood pressure change trend; When the predicted blood pressure parameters exceed the static alarm threshold or the relative change threshold, a blood pressure abnormality warning is triggered.

[0036] In one feasible implementation, the above-mentioned safety arbitration and anomaly protection control based on multimodal pressure signals to complete the safety management of the blood pressure measurement process includes: The cuff inflation pressure signal and the cuff interface pressure signal are collected in real time, and the current cuff pressure value corresponding to the above cuff inflation pressure signal is displayed; at the same time, the above cuff inflation pressure signal and the above cuff interface pressure signal are compared in real time, and the pressure difference between the two is calculated. When the pressure difference exceeds the preset difference threshold, or when the cuff pressure exceeds the preset maximum pressure threshold and the duration exceeds the set duration, it is determined to be an abnormal state. In the above-mentioned abnormal state, safety control operations are performed, including at least triggering the pressure protection fuse mechanism, terminating inflation, opening the pressure relief valve for automatic pressure relief, and stopping the current measurement process.

[0037] For example, after completing the acquisition and quality assessment of multimodal physiological signals, adaptive weighted fusion is first performed on the signals of different modalities to improve the stability and anti-interference ability of subsequent blood pressure calculation. Since the reliability of cuff inflation pressure signals, cuff interface pressure signals, pulse wave signals, and ECG signals varies under different patient states, wearing conditions, and interference environments, this embodiment does not use a fixed-weight fusion method. Instead, it dynamically adjusts the contribution of each modality to blood pressure calculation based on its real-time quality level. Specifically, the first... The quality score of each modal signal is denoted as . And construct the following adaptive fusion weight allocation formula:

[0038] in, Indicates the first The fusion weights corresponding to each modal signal Indicates the first Quality score of each modal signal This represents the total number of modes participating in the fusion. This represents the weight sensitivity coefficient, used to control the degree to which high-quality modes dominate the final fusion result. When As the weight increases, higher-quality modes will receive greater fusion weights. Based on these weights, the fused signal can be further obtained:

[0039] in, Indicates time The fused signal Indicates the first The modality at time... The original or preprocessed signal value.

[0040] Using the above method, when the quality of a certain modality deteriorates due to motion artifacts, cuff detachment, or poor sensor contact, its weight will be automatically reduced, while the weight of the higher-quality modality will be enhanced, thus ensuring that the fused signal still has high reliability. Especially when the quality of the first modality (cuff inflation pressure signal) is below a threshold while the quality of the second modality (cuff interface pressure signal) remains high, the system can set the second modality as the dominant signal to avoid misjudgments or measurement failures caused by abnormalities in the main pressure channel in traditional single-channel oscilloscope measurements. If all modal signals are below the preset quality threshold, it indicates that the current measurement environment cannot support reliable calculation. In this case, the system directly terminates the current measurement process and outputs a signal abnormality warning to prevent invalid or erroneous results from entering the subsequent blood pressure estimation stage.

[0041] After obtaining the fused signal, this embodiment further extracts features and performs temporal modeling on the oscillatory wave envelope in the fused signal based on a deep learning model. Traditional oscillatory wave methods typically rely on fixed empirical ratios to determine the positions of systolic and diastolic blood pressure. However, this method is difficult to adapt to individual differences among patients and is easily affected by arrhythmias, waveform distortion, and noise interference. Therefore, this embodiment uses a convolutional neural network to extract local morphological features and a long short-term memory network to model the temporal change patterns for intelligent identification of the oscillatory wave envelope. First, morphological parameters such as the slope of the rising branch, the slope of the falling branch, the peak pressure point, the envelope width, and waveform asymmetry are extracted from the oscillatory wave envelope and constructed as oscillatory wave feature vectors. Subsequently, the feature vectors from multiple consecutive cardiac cycles are input into the long short-term memory network to learn the dynamic evolution of the oscillatory wave throughout the deflation process, thereby outputting blood pressure parameter prediction results. To improve the ability to identify abnormal waveforms, the following abnormal oscillatory wave scoring formula is constructed:

[0042] in, Indicates the score of abnormal oscillation wave. This indicates the slope of the rising branch of the current oscillating wave. This indicates the slope of the descending branch of the current oscillating wave. This indicates the current oscillating wave envelope width. Indicates the normal reference envelope width. This indicates the pressure point corresponding to the current peak position. Indicates the reference peak position. This represents the weighting coefficients corresponding to each abnormal feature, used to adjust the contribution of different abnormal factors to the total score. When When the current oscillation waveform exceeds a preset threshold, it is determined that there is an anomaly. The waveform can then be removed, downweighted, or reconstructed to prevent abnormal heartbeats or transient interference from directly affecting the final blood pressure prediction result. This method enhances the system's waveform recognition capabilities in scenarios involving atrial fibrillation, premature beats, and partial cuff slippage.

[0043] After completing the initial blood pressure prediction, this embodiment further introduces an individualized blood pressure baseline model based on historical measurement data to correct for deviations in the current prediction results. The aim is to address the problem that traditional non-invasive blood pressure monitoring relies solely on single absolute values, making it difficult to reflect individual patient differences. Specifically, the system collects historical blood pressure measurement data and corresponding pulse wave, electrocardiogram, and pressure characteristic data within a preset time window. Outliers caused by significant motion interference, waveform defects, or equipment malfunctions are removed and cleaned. Based on this, the patient's baseline systolic blood pressure, baseline diastolic blood pressure, and baseline heart rate are calculated. To ensure that the baseline model retains historically stable characteristics while reflecting recent physiological changes, the following individualized baseline update formula can be constructed:

[0044] in, Indicates the first The updated individualized baseline blood pressure. This indicates the baseline blood pressure from the previous reading. recen represents the mean of valid blood pressure measurements within the most recent time window. This represents the historical baseline preservation coefficient, and its value range is typically [value range missing]. , The larger the value, the more the system values ​​its historical stable state. The smaller the value, the more sensitive the system is to recent blood pressure changes. Based on this baseline model, the system can further correct for deviations in the current blood pressure prediction results, so that the output not only reflects the calculated value of the current oscillation wave, but also considers the patient's long-term blood pressure distribution characteristics and short-term physiological trends, thereby improving the adaptability to individual patients.

[0045] During safety management, a dual-channel pressure monitoring mechanism is established using both cuff inflation pressure and cuff interface pressure signals. The cuff inflation pressure signal is acquired in real-time by a pressure sensor located within the cuff's air path, characterizing the actual air pressure applied by the inflation pump to the cuff bladder. The cuff interface pressure signal is acquired in real-time by a pressure sensor located on the inside of the cuff or at the point of contact between the cuff and the body, characterizing the actual pressure level exerted by the cuff on the limb surface. The system continuously acquires these two pressure signals during blood pressure measurement and sends the current cuff pressure value corresponding to the inflation pressure signal to the display module. This value is displayed in real-time as a numerical value, pressure level, pressure change curve, or other visual representation, allowing medical personnel or users to intuitively understand the current cuff pressure status and pressure change process.

[0046] Furthermore, while displaying the cuff pressure level in real time, the system continuously verifies the consistency between the cuff inflation pressure signal and the cuff interface pressure signal. Specifically, it calculates the pressure difference between the cuff inflation pressure value and the cuff interface pressure value at the same sampling moment and compares this pressure difference with a preset difference threshold. Under normal inflation and measurement conditions, the air pressure can be effectively transmitted to the human limb through the cuff airbag, so the two pressure signals usually maintain a relatively stable correspondence. When the cuff becomes loose, the wearing position is off, the air path is blocked, the airbag malfunctions, or the pressure sensor fails, a significant deviation may occur between the two pressure signals. Therefore, when the detected pressure difference exceeds the preset difference threshold, it can be considered that there is an abnormality in the current cuff pressurization state or pressure detection state.

[0047] This embodiment also monitors the absolute pressure level of the cuff itself and the duration of the high-pressure state. Specifically, the real-time cuff pressure value is compared with a preset maximum pressure threshold, and a timer is started after the cuff pressure value exceeds the preset maximum pressure threshold; when the duration of the overpressure state exceeds a set time, it is determined that there is a risk of continuous over-pressurization in the current measurement process. Therefore, the system does not only rely on the consistency between the two pressure signals for safety judgment. Even if the cuff inflation pressure signal and the cuff interface pressure signal are consistent, when the actual pressure corresponding to the two has reached an excessively high level and lasts for a long time, abnormal protection can still be triggered, thus forming a pressure double insurance mechanism composed of "pressure consistency anomaly detection" and "absolute pressure over-limit and timeout detection".

[0048] When any of the above-mentioned abnormal judgment conditions are met, the system enters an abnormal state and triggers the pressure protection fuse mechanism. The pressure protection fuse mechanism can be triggered by the safety management unit sending a high-priority safety control command to the inflation drive module and the pressure relief execution module, causing the inflation pump to stop working or cut off the inflation drive signal, while simultaneously opening the pressure relief valve to rapidly expel gas from the cuff, thereby reducing the cuff pressure and stopping the current blood pressure measurement process. After the fuse protection is triggered, the system can maintain a state where re-inflation is prohibited until the cuff pressure is detected to have dropped to a safe pressure range or the abnormal state is cleared, to prevent the cuff from inflating again due to control program abnormalities, sensor malfunctions, or continuous inflation commands.

[0049] In this embodiment, the real-time visualization of cuff pressure, dual-channel pressure consistency judgment, maximum pressure limit, overpressure duration limit, and automatic fuse depressurization are combined to enable the system to detect pressure mismatch caused by abnormalities in the gas path or sensor, and to stop inflation and automatically depressurize in time when the actual cuff pressure is too high even though there is no obvious mismatch in the pressure signal itself. This prevents the cuff from maintaining excessively high pressure for a long time and improves the safety and reliability of abnormal protection during non-invasive blood pressure measurement.

[0050] In summary, this invention fundamentally improves the traditional non-invasive blood pressure monitoring technology by introducing multimodal physiological signal fusion and artificial intelligence processing mechanisms. By simultaneously acquiring cuff inflation pressure, cuff interface pressure, pulse wave, and ECG signals, a multi-dimensional sensing system is constructed. Adaptive weighted fusion based on signal quality assessment ensures the system can still output stable and reliable blood pressure results even when a single signal is interfered with or fails. By employing a deep learning model to extract features and perform temporal modeling of the oscillatory wave envelope, the invention effectively overcomes the problem of insufficient adaptability to individual differences in traditional fixed threshold methods, improving the accuracy of blood pressure calculation, especially maintaining high precision in scenarios with arrhythmias or complex interference. This invention constructs an individualized blood pressure baseline model based on historical measurement data and dynamically generates alarm thresholds by combining short-term trend analysis, realizing a shift from absolute threshold monitoring to relative change monitoring. This allows for early identification of abnormal blood pressure changes and enhances clinical early warning capabilities. Through real-time comparison of dual-modal pressure signals and a safety arbitration mechanism, protective operations are promptly executed when abnormal pressure or sensor malfunction is detected, effectively avoiding safety risks such as continuous inflation, and switching to a degraded monitoring mode when necessary to ensure monitoring continuity.

[0051] Secondly, this invention also proposes a non-invasive cuff-type blood pressure monitoring system based on artificial intelligence, such as... Figure 2 As shown, the method for implementing the artificial intelligence-based non-invasive cuff blood pressure monitoring method according to any one of the first aspects of claim includes: The acquisition unit 21 is used to acquire multimodal physiological signal data, wherein the multimodal physiological signal data includes at least cuff inflation pressure signal, cuff interface pressure signal, pulse wave signal and electrocardiogram signal; Evaluation unit 22 is used to evaluate the signal quality of the above multimodal physiological signal data and obtain the quality level of each modality signal; The fusion unit 23 is used to adaptively weight and fuse the multimodal physiological signal data based on the above signal quality level to obtain a fused signal; Modeling unit 24 is used to extract features and perform time-series modeling of the oscillating wave envelope in the above fused signal based on a deep learning model, so as to obtain blood pressure parameter prediction results; The correction unit 25 is used to construct an individualized blood pressure baseline model based on historical measurement data, and to correct the blood pressure parameter prediction results based on the baseline model. The generation unit 26 is used to dynamically generate alarm thresholds based on the above blood pressure parameter prediction results and the above baseline model, and to perform blood pressure abnormality warning. Safety management unit 27 is used to perform safety arbitration and abnormal protection control based on multimodal pressure signals to complete the safety management of the blood pressure measurement process.

[0052] In one feasible implementation, an AI-based non-invasive cuff blood pressure monitoring system can also perform any step of the method proposed in the first aspect.

[0053] Thirdly, the present invention also proposes an electronic device 300, such as... Figure 3 As shown, it includes a memory 310, a processor 320, and a computer program 311 stored on the memory 310 and executable on the processor. When the processor 320 executes the computer program 311, it implements the steps of the artificial intelligence-based non-invasive cuff blood pressure monitoring method as described in any of the first aspects.

[0054] Fourthly, the present invention also proposes a computer-readable storage medium having a computer program stored thereon, characterized in that, when the computer program is executed by a processor, it implements the steps of the artificial intelligence-based non-invasive cuff blood pressure monitoring method as described in any one of the first aspects.

[0055] It should be noted that the descriptions of each embodiment in the above embodiments have different focuses. For parts that are not described in detail in a certain embodiment, please refer to the relevant descriptions in other embodiments.

[0056] The above embodiments are only used to illustrate the technical solutions of this application, and are not intended to limit them. Although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of this application.

Claims

1. A non-invasive cuff-type blood pressure monitoring method based on artificial intelligence, characterized in that, include: Acquire multimodal physiological signal data, wherein the multimodal physiological signal data includes at least cuff inflation pressure signal, cuff interface pressure signal, pulse wave signal and electrocardiogram signal; The multimodal physiological signal data are subjected to signal quality assessment to obtain the quality level of each modality signal; Based on the signal quality level, the multimodal physiological signal data are adaptively weighted and fused to obtain a fused signal; Based on a deep learning model, feature extraction and temporal modeling are performed on the oscillatory wave envelope in the fused signal to obtain blood pressure parameter prediction results. An individualized blood pressure baseline model is constructed based on historical measurement data, and the blood pressure parameter prediction results are corrected according to the baseline model. Based on the blood pressure parameter prediction results and the baseline model, alarm thresholds are dynamically generated, and abnormal blood pressure warnings are executed. Safety arbitration and anomaly protection control are performed based on multimodal pressure signals to complete the safety management of the blood pressure measurement process.

2. The non-invasive cuff-type blood pressure monitoring method based on artificial intelligence as described in claim 1, characterized in that, The process of evaluating the signal quality of the multimodal physiological signal data to obtain the quality level of each modality signal includes: Quality assessment features of each modality of physiological signal are extracted, including signal amplitude stability, signal-to-noise ratio, waveform integrity and motion interference index. Construct signal quality assessment indicators based on the aforementioned quality evaluation characteristics; The signal quality evaluation index is compared with a preset quality threshold to obtain the corresponding signal quality level.

3. The non-invasive cuff-type blood pressure monitoring method based on artificial intelligence as described in claim 1, characterized in that, The adaptive weighted fusion of the multimodal physiological signal data based on the signal quality level to obtain the fused signal includes: Assign corresponding fusion weights based on the quality level of each modal signal; The initial fused signal is obtained by weighting each modal signal based on the fusion weights. When the quality of the first modal signal is detected to be lower than a preset threshold and the quality of the second modal signal is higher than a preset threshold, the fusion weights are adjusted so that the second modal signal becomes the dominant signal. When the quality of all multimodal signals is below a preset threshold, the current measurement is terminated and a signal abnormality prompt is output.

4. The non-invasive cuff-type blood pressure monitoring method based on artificial intelligence as described in claim 1, characterized in that, The deep learning model is used to extract features and perform temporal modeling on the oscillatory wave envelope in the fused signal to obtain blood pressure parameter prediction results, including... The morphological features of the oscillating wave envelope are extracted based on a convolutional neural network to obtain an oscillating wave feature vector, wherein the oscillating wave feature vector includes the rising branch slope, the falling branch slope, the peak position, the envelope width, and the asymmetry index. The oscillation wave feature vector is modeled temporally using a long short-term memory network to obtain the dynamic change law of the oscillation wave, and abnormal oscillation waveforms are identified and eliminated to obtain blood pressure parameter prediction results.

5. The non-invasive cuff-type blood pressure monitoring method based on artificial intelligence as described in claim 1, characterized in that, The process of constructing an individualized blood pressure baseline model based on historical measurement data, and correcting the predicted blood pressure parameters based on the baseline model, includes: Acquire historical blood pressure measurement data and corresponding multimodal feature data within a preset time window; The historical measurement data were subjected to outlier removal and data cleaning. Calculate the baseline values ​​for systolic blood pressure, diastolic blood pressure, and heart rate; An individualized blood pressure baseline model is constructed based on the baseline values; The bias correction is performed on the current blood pressure parameter prediction results based on the individualized blood pressure baseline model.

6. The non-invasive cuff-type blood pressure monitoring method based on artificial intelligence as described in claim 1, characterized in that, The step of dynamically generating alarm thresholds based on the blood pressure parameter prediction results and the baseline model, and executing blood pressure abnormality warnings, includes: The static alarm threshold is calculated based on the individualized blood pressure baseline model, wherein the static alarm threshold includes an upper limit threshold for systolic blood pressure and a lower limit threshold for systolic blood pressure; Based on the blood pressure parameter prediction results, a short-term blood pressure change trend model is constructed to predict the blood pressure change trend within a preset time period. Calculate the relative change threshold based on the described blood pressure change trend; When the predicted blood pressure parameter exceeds the static alarm threshold or the relative change threshold, a blood pressure abnormality warning is triggered.

7. The non-invasive cuff-type blood pressure monitoring method based on artificial intelligence as described in claim 1, characterized in that, The safety arbitration and anomaly protection control based on multimodal pressure signals, to complete the safety management of the blood pressure measurement process, includes: The cuff inflation pressure signal and the cuff interface pressure signal are collected in real time, and the current cuff pressure value corresponding to the cuff inflation pressure signal is displayed; at the same time, the cuff inflation pressure signal and the cuff interface pressure signal are compared in real time, and the pressure difference between the two is calculated. When the pressure difference exceeds a preset difference threshold, or when the cuff pressure exceeds a preset maximum pressure threshold and the duration exceeds a set time, it is determined to be an abnormal state. In the abnormal state, a safety control operation is performed, which includes at least triggering the pressure protection fuse mechanism, terminating inflation, opening the pressure relief valve for automatic pressure relief, and stopping the current measurement process.

8. A non-invasive cuff-type blood pressure monitoring system based on artificial intelligence, used to implement the non-invasive cuff-type blood pressure monitoring method based on artificial intelligence as described in any one of claims 1 to 7, comprising: The acquisition unit is used to acquire multimodal physiological signal data, wherein the multimodal physiological signal data includes at least cuff inflation pressure signal, cuff interface pressure signal, pulse wave signal and electrocardiogram signal; An evaluation unit is used to evaluate the signal quality of the multimodal physiological signal data and obtain the quality level of each modality signal. The fusion unit is used to adaptively weight and fuse the multimodal physiological signal data based on the signal quality level to obtain a fused signal; The modeling unit is used to extract features and perform temporal modeling on the oscillatory wave envelope in the fused signal based on a deep learning model, so as to obtain blood pressure parameter prediction results. The calibration unit is used to construct an individualized blood pressure baseline model based on historical measurement data, and to correct the blood pressure parameter prediction results according to the baseline model; The generation unit is used to dynamically generate alarm thresholds based on the blood pressure parameter prediction results and the baseline model, and to perform blood pressure abnormality warnings. The safety management unit is used to perform safety arbitration and anomaly protection control based on multimodal pressure signals to complete the safety management of the blood pressure measurement process.

9. An electronic device, comprising: The memory and processor are characterized in that the processor is used to execute a computer program stored in the memory to implement the steps of the artificial intelligence-based non-invasive cuff blood pressure monitoring method as described in any one of claims 1-7.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it implements the steps of the artificial intelligence-based non-invasive cuff blood pressure monitoring method as described in any one of claims 1-7.

Images